Interactive comment on “ Size-resolved measurements of ice nucleating particles at six locations in North America and one in Europe ”

Abstract. Detailed information on the size of ice-nucleating particles (INPs) may be useful in source identification, modeling their transport in the atmosphere to improve climate predictions, and determining how effectively or ineffectively instrumentation used for quantifying INPs in the atmosphere captures the full INP population. In this study we report immersion-mode INP number concentrations as a function of size at six ground sites in North America and one in Europe using the micro-orifice uniform-deposit impactor droplet freezing technique (MOUDI-DFT), which combines particle size-segregation by inertial impaction and a microscope-based immersion freezing apparatus. The lowest INP number concentrations were observed at Arctic and alpine locations and the highest at suburban and agricultural locations, consistent with previous studies of INP concentrations in similar environments. We found that 91 ± 9, 79 ± 17, and 63 ± 21 % of INPs had an aerodynamic diameter > 1 µm at ice activation temperatures of −15, −20, and −25 °C, respectively, when averaging over all sampling locations. In addition, 62 ± 20, 55 ± 18, and 42 ± 17 % of INPs were in the coarse mode (> 2.5 µm) at ice activation temperatures of −15, −20, and −25 °C, respectively, when averaging over all sampling locations. These results are consistent with six out of the nine studies in the literature that have focused on the size distribution of INPs in the atmosphere. Taken together, these findings strongly suggest that supermicron and coarse-mode aerosol particles are a significant component of the INP population in many different ground-level environments. Further size-resolved studies of INPs as a function of altitude are required since the size distribution of INPs may be different at high altitudes due to size-dependent removal processes of atmospheric particles.


Introduction
Ice-nucleating particles (INPs) are a unique class of aerosol particles that catalyze ice formation under atmospheric conditions.A variety of particle types have been identified as INPs, including mineral dust, black carbon, volcanic ash, glassy aerosols, and primary biological particles such as bacteria, fungal spores, and pollen (see reviews by Szyrmer and Zawadzki, 1997;Möhler et al., 2007;Ariya et al., 2009;Després et al., 2012;Hoose and Möhler, 2012;Murray et al., 2012;Yakobi-Hancock et al., 2013).Although only a small fraction of aerosol particles nucleate ice (e.g., Rogers et al., 1998), INPs are important since they can lead to changes in the properties and lifetimes of mixed-phase and ice clouds, R. H. Mason et al.: Size-resolved measurements of ice-nucleating particles ultimately affecting climate and precipitation (Baker, 1997;Lohmann and Feichter, 2005;Baker and Peter, 2008;DeMott et al., 2010;Creamean et al., 2013).Vali et al. (2015) describes four modes of heterogeneous ice nucleation: deposition nucleation, where ice forms on the INP directly from the gas phase; condensation freezing, where ice nucleates during the condensing of water onto the INP; immersion freezing, where crystallization is initiated by an INP within a supercooled liquid droplet; and contact freezing, where the freezing of a supercooled liquid droplet is due to impaction by an INP.In this study we focus on freezing via the immersion mode in dilute solution droplets, which is relevant to mixed-phase cloud conditions.
Due to the importance of INPs for climate and precipitation, there has been a renewed interest in measuring the concentrations of INPs in the atmosphere (DeMott et al., 2011).While much of this work has focused on measurements of the total number concentration of INPs, there has been less emphasis on determining their size distributions in the atmosphere.Information on airborne INP size distributions may be particularly helpful in identifying the predominant INP sources.For example, information on the size distribution of INPs may help rule out or support the role of fungal spores in atmospheric ice nucleation since they are often in the supermicron range (Graham et al., 2003;Elbert et al., 2007;Sesartic and Dallafior, 2011;Després et al., 2012;Huffman et al., 2012).A similar approach can be used with black carbon particles, since they are mainly in the submicron range (Clarke et al., 2004;Schwarz et al., 2008Schwarz et al., , 2013)).
Previous modeling studies have shown that the transport and distribution of INPs, and aerosol particles in general, are sensitive to the size of the particles assumed in the models (Burrows et al., 2009;Wilkinson et al., 2011).Information on the size distributions of INPs are thus needed for accurate modeling of their transport and distributions in the atmosphere (Morris et al., 2004;Hoose et al., 2010a, b;Sesartic et al., 2013;Haga et al., 2014;Spracklen and Heald, 2014).
Information on the size distribution of INPs is also needed to determine whether techniques used to measure atmospheric INP concentrations capture the entire INP population.For example, the continuous flow diffusion chamber (Rogers et al., 2001b) is often used for measuring INPs (e.g., DeMott et al., 1998;Rogers et al., 2001a;Richardson et al., 2007;Pratt et al., 2009;Prenni et al., 2009;Eidhammer et al., 2010;Chou et al., 2011;Friedman et al., 2011;Hoyle et al., 2011;Corbin et al., 2012;Garcia et al., 2012;Tobo et al., 2013;McCluskey et al., 2014).This type of instrument has the advantage of providing real-time measurements of INPs with the ability to detect very large INP number concentrations, but the aerodynamic diameter of particles measured with it is limited, from d 50 ≤ 2.4 µm in some studies (e.g., Garcia et al., 2012) to d 50 ≤ 0.75 µm in others (e.g., De-Mott et al., 2003).Such techniques may miss supermicron or coarse-mode (i.e., larger than 2.5 µm) INPs.The exact proportion of INPs missed may depend on temperature.Such online instruments have typically focused on measurements below approximately −20 • C as sample volume considerations limit effective sampling of lower INP number concentrations at warmer temperatures.The exact proportion of INPs missed may also depend on altitude since the removal of atmospheric particles by wet and dry deposition in the atmosphere is expected to be size dependent.As an example, supermicron particles have larger dry deposition loss rates than submicron particles.
Previous studies of INPs as a function of size have been carried out in the field (e.g., Vali, 1966;Rosinski et al., 1986;Mertes et al., 2007;Santachiara et al., 2010) and in the laboratory (e.g., Welti et al., 2009;O'Sullivan et al., 2015).These and additional studies are further discussed in Sect.3.2.In the current study, we add to the existing body of size-resolved INP measurements by reporting ground-level INP size distributions from six locations in North America and one in Europe, covering a diverse set of environments and investigating immersion freezing at −15, −20, and −25 • C.

Sampling sites
The seven locations used in this study are detailed in Table 1 and shown in Fig. 1.All reported sampling periods are local times.Measurements using the sampling instrumentation described in the next section were made at five locations in Canada: Alert, Nunavut; the Labrador Sea near Newfoundland and Labrador; Whistler Mountain, British Columbia; the University of British Columbia (UBC) campus, British Columbia; and Amphitrite Point, British Columbia.Measurements in Canada were conducted as part of the larger NETwork on Climate and Aerosols: addressing key uncertainties in Remote Canadian Environments project (NET-CARE; http://netcare-project.ca/).Measurements were also made at Saclay, France and Colby, Kansas, USA.

Alert
Arctic sampling was conducted at the Dr. Neil Trivett Global Atmosphere Watch Observatory in Alert, Nunavut, Canada (labeled 1 in Fig. 1; Cobbett et al., 2007) between 29 March and 23 July 2014.This Arctic research station is part of a global network for measuring chemical and physical perturbations of the atmosphere.Aerosol particles were collected through a louvered total suspended particulate (TSP) inlet (Mesa Labs Inc., Butler, NJ, USA) and 0.9 m mast located on the upper level of an outdoor platform free of surrounding obstructions, and were stored in the dark at −15 or 4 • C for a period of 10-112 days prior to analysis.

Whistler Mountain
The Whistler Peak High Elevation Site is located at the summit of Whistler Mountain in Whistler, British Columbia, Canada (labeled 2 in Fig. 1) and operated by Environment Canada (Gallagher et al., 2011;Macdonald et al., 2011).mobile operation.Samples were stored in the dark at 4 • C for a period of 1-4 days prior to analysis.

Amphitrite Point
The coastal site at Amphitrite Point on Vancouver Island, British Columbia, Canada (labeled 3 in Fig. 1) is operated by Environment Canada, the BC Ministry of Environment, and Metro Vancouver for the continuous monitoring of aerosols and trace gases influenced by marine trajectories (McKendry et al., 2014;Yakobi-Hancock et al., 2014;Mason et al., 2015b).The mobile laboratory used during sampling was located approximately 100 m from the high tide line of the Pacific Ocean along a rocky shoreline, separated from the ocean by a narrow row of trees and shrubs approximately 2-10 m in height.Sampling took place from 6 to 27 August 2013 using a louvered TSP inlet and 3 m mast.Aerosol particles were stored at room temperature and analyzed within 1 day of collection.

The Labrador Sea
The Canadian Coast Guard Service vessel CCGS Amundsen serves as both an icebreaker for shipping lanes and an Arctic research vessel.One set of aerosol particle samples was collected from the top of the bridge of this vessel on July 11, 2014 while in the Labrador Sea off the coast of Newfoundland and Labrador, Canada (labeled 4 in Fig. 1).While sampling was within the marine boundary layer in the presence of sea spray aerosols, back trajectories (not included) calculated using the Hybrid Single-Particle Lagrangian Integrated Trajectory (HYSPLIT4) model of the National Oceanographic and Atmospheric Administration (Draxler and Rolph, 2014) indicate that the sampled air mass spent the majority of the previous 72 h period over land.Air was passed through a louvered TSP inlet and 1.5 m mast during sampling, and collected aerosol particles were stored in the dark at 4 • C for a period of 45-46 days prior to analysis.

Saclay, France
Aerosol particle samples were collected at the Commissariat à l'Energie Atomique (CEA) Atmospheric Supersite (AS), CEA l'Orme des Merisiers.The CEA-AS Observatory is a suburban area located 30 km southeast of Paris in Saclay, France (labeled 5 in Fig. 1).The CEA-AS Observatory is surrounded by different sources of bioaerosols such as forest and agricultural fields, and is often influenced by marine or urban air masses (Baisnée et al., 2014).Measurements were made as part of the BIODETECT 2014 intensive campaign, an intercomparison of bioaerosol detection methods (Sarda-Estève et al., 2014).During this study period, the site was heavily influenced by urban outflow.A large set of ancillary measurements was done to constrain all the particulate matter sources.Aerosol particles were sampled through a TSP inlet and 10 m mast between 15 July and 4 August 2014, and were stored in the dark at 4 • C for a period of 55-217 days prior to analysis.

UBC campus
Four sets of aerosol particle samples were collected from a weather station on the roof of the five-story Earth Sciences Building on the UBC campus in British Columbia, Canada (labeled 6 in Fig. 1).The UBC campus is located on a peninsula and is surrounded by forest on three sides and ocean on the fourth.The site has been classified as suburban since it is less than 10 km from downtown Vancouver.Samples were collected through a TSP inlet and 0.5 m mast between 12 and 16 May 2014.The aerosol particles were stored in the dark at 4 • C for a period of 21-23 days prior to analysis.

Colby, KS
Aerosol particles were collected at the soybean and sorghum fields of the Kansas State University Northwest Research Center in Colby, KS, USA (labeled 7 in Fig. 1).One sample was collected at each location during combine harvesting from a distance approximately 3-10 m downwind of the field.A third sample was also collected at the sorghum field the night following harvest.Sampling took place on 14 and 15 October 2014 and samples were stored in the dark at 4 • C for a period of 41-46 days prior to analysis.

Size-resolved INP number concentrations
INP number concentrations as a function of size and temperature were determined using the micro-orifice uniformdeposit impactor droplet freezing technique (MOUDI-DFT; Huffman et al., 2013;Mason et al., 2015a).This technique combines aerosol particle collection by a cascade inertial impactor with sharp size-cutoff characteristics (the MOUDI; Marple et al., 1991) with an established droplet freezing apparatus (the DFT) for determining immersion-mode freezing properties (Koop et al., 1998;Iannone et al., 2011;Haga et al., 2013).A similar approach has also been used to study deposition nucleation by particles collected from the atmosphere (Wang et al., 2012;Knopf et al., 2014).
Bounce within inertial impactors such as the MOUDI can occur during aerosol sampling, where particles impact the collection substrate but are not retained.This rebounding of particles from the surface could possibly alter the INP number concentrations and size distributions being measured.If composition is held constant, bounce is expected to increase with particle size because of their greater kinetic energy (Dahneke, 1971).Hence, INP number concentrations for large particle sizes may be underestimated here.Bounce is also expected to increase with decreasing relative humidity (RH).Previous work has shown that having a sample RH of 70 % or greater can be effective in reducing particle bounce (e.g., Winkler, 1974;Fang et al., 1991;Stein et al., 1994;Vasiliou et al., 1999;Chen et al., 2011;Bateman et al., 2014), although its efficacy is dependent on particle type (Winkler, 1974;Lawson, 1980;Saukko et al., 2012).For six out of the seven sites investigated here, the average RH during sampling was 69 % or greater.Recently, results from the MOUDI-DFT and the continuous flow diffusion chamber were compared during an ambient field campaign at Colorado State University (Mason et al., 2015a).For particle sizes < 2.4 µm, the INP number concentrations measured by the MOUDI-DFT were within experimental error of those measured by the continuous flow diffusion chamber technique, suggesting that bounce was not an issue during these previous ambient field measurements.Based on these previous measurements, in the current studies we do not consider the issue of particle bounce when calculating INP number concentrations and size distributions.Nevertheless, additional studies are warranted to better quantify the effect of bounce.
During sampling, the aerosols were not brought to a standardized humidity, and the RH varied from site-to-site (see Table 1 for the average RH during sampling at each site).This variability in RH could lead to a small variability in INP size due to differences in hygroscopic growth.We further note that the duration of sample storage varied in the current study.Additional studies are needed to quantify the effect of storage on INP activity.

Freezing measurements
Samples were analyzed by the DFT to determine the number concentration of particles active in the immersion-freezing mode.Details of the experimental procedure can be found in Mason et al. (2015a).Briefly, samples were transferred to a temperature-and humidity-controlled flow cell coupled to an optical microscope equipped with a 5× magnification objective (Axiolab; Zeiss, Oberkochen, Germany).Water droplets were condensed onto the sample and monitored using a CCD camera recording a digital video.Since the relative humidity of the gas flow during droplet condensation was held at approximately 120 %, water condensation occurred uniformly on the cover slip, and growing droplets coagulated as they grew to a final size of 97 ± 42 µm (mean diameter and 1 standard deviation (SD) uncertainty).The freezing temperature of each droplet was then determined during cooling at a rate of −10 • C min −1 using the video timestamp and a resistance temperature detector located within the flow cell.Note that droplet growth and coagulation occurs in the same manner for samples containing particles and clean hydrophobic glass cover slips (no particles deposited).In addition, based on an analysis of samples collected at Amphitrite Point, more than 99 % of particles become incorporated into the droplets prior to the freezing experiments.Here we regard ice nucleation as a singular process (i.e., strictly temperature-dependent) but note that the stochastic (i.e., time-dependent) component to immersion freezing (Vali, 2014) may alter the median freezing temperature of a droplet by 0.5-2 • C per decade change in cooling rate (Murray et al., 2011;Welti et al., 2012;Wright and Petters, 2013;Wright et al., 2013;Wheeler et al., 2015).
A potential issue with the droplet freezing technique is heterogeneous ice nucleation initiated by the hydrophobic glass cover slips used to collect atmospheric particles.To address this issue, experiments were conducted using new hydrophobic glass cover slips that were processed in the same manner as ambient samples except they were not exposed to atmospheric particles drawn into the MOUDI.For the five hydrophobic glass cover slips investigated, which contained 231 droplets generated during the freezing experiments, the average freezing temperature was −36.5 ± 0.5 • C (1 SD).In addition, none of the droplets froze above −33.7 • C in these blank experiments.Since we only report INP number concentrations for temperatures from −15 to −25 • C in this study, heterogeneous ice nucleation by the substrate is unlikely to contribute to the reported INP number concentrations.

Calculating the number concentration of INPs
The number of INPs in the DFT, #INPs(T ), was calculated using the following equation which accounts for the possibility of a droplet containing multiple INPs (Vali, 1971): where N u (T ) is the number of unfrozen droplets at temperature T , N o is the total number of droplets, f nu,0.25−0.10mm is a non-uniformity correction factor that takes into account nonuniformity at the 0.25-0.10mm scale, and f ne is a statistical uncertainty derived for a given number of detected nucleation events, with fewer nucleation events leading to greater statistical uncertainty (Koop et al., 1997).For the results reported, here f ne is derived for a confidence level of 0.95.
Equation (1) assumes that droplets in a given freezing experiment have the same volume (Vali, 1971).Using droplet freezing experiments reported in Mason et al. (2015a), which are similar to experiments presented here, we explored whether this assumption leads to uncertainties when applied to the DFT experiments.Number of INPs was first calculated from the DFT experiments as described above.Second, number of INPs were calculated by first separating the droplet freezing results into 2-4 bins based on droplet volume.After binning the data by droplet volume, droplets in each bin are more similar in volume.Equation (1) was then used to calculate the number of INPs in each bin.Finally, the total number of INPs was determined by summing the numbers of INPs calculated for each bin.We found that the total number of INPs determined both ways (with and without binning) agreed within the experimental uncertainty at freezing temperatures of −25 • C and above for 97 % of freezing events.We also analyzed the same data by first separating the droplet freezing results into 2-4 bins based on the maximum area the droplets covered.Again, the number of INPs determined with and without binning was in good agreement, being within the experimental uncertainty at freezing temperatures of −25 • C and above for 98 % of freezing events.Based on this analysis we conclude that the application of Eq. ( 1), which assumes monodisperse droplets, to the DFT results at −25 • C and above does not lead to large uncertainties.
In the DFT, once a droplet freezes it may grow by vapor diffusion and contact a neighboring liquid droplet, causing the latter to freeze as well.To account for these nonimmersion freezing events, we calculated the #INPs(T ) in the immersion mode using two different scenarios: (i) we calculated an upper limit by assuming that all droplets which underwent the processes discussed above froze by immersion freezing; and (ii) we calculated a lower limit to the INP concentration by assuming that all droplets which underwent the processes discussed above remained liquid until the homogeneous freezing temperature of approximately −37 • C  (Wheeler et al., 2015).
After #INPs(T ) were determined for a given freezing experiment, atmospheric INP number concentrations, [INPs(T )], were calculated using the following equation: where A deposit is the total area of the sample deposit on the MOUDI impaction plate, A DFT is the area of the sample analyzed by the DFT (1.2 mm 2 in all samples), V is the volume of air sampled by the MOUDI, and f nu,1 mm is a correction factor to account for non-uniformity in particle concentration of the sample deposit at the 1 mm scale.See Mason et al. (2015a) for details.Values of A deposit , f nu,0.25−0.10mm , f nu,1 mm , and f ne are given in Tables S1 and S2 and discussed in Sect.S1 of the Supplement.Reported INP number concentrations at each location are averaged over all samples and have been adjusted to standard temperature and pressure.In calculating averages over all sampling locations, measurements have not been weighted by sample number.
For the experimental conditions used in the current study the maximum number concentration of INPs that could be detected was roughly 20 L −1 .This maximum number concentration is greater than that reported in Huffman et al. (2013) because in the current studies shorter sampling times were used.

INP number concentrations
The total number concentration of INPs active at −15, −20, and −25 • C are shown for each site in Fig. 2. Freezing events were rare at temperatures warmer than −15 • C, accounting for only 1.3 % of all cases, and are therefore not reported.Some of the DFT experiments proceeded such that all droplets froze at temperatures slightly below −25 • C. Since this scenario prohibits calculation of INP number concentrations, −25 • C is the lowest temperature reported.As expected, INP number concentrations were found to increase with decreasing freezing temperature with the average concentration at −25 • C (3.8 ± 2.9 L −1 ) being more than an order of magnitude larger than at −15 • C (0.25 ± 0.15 L −1 ).
INP number concentrations were relatively low at the Alert and Whistler Mountain sites with values of 0.05 and 0.10 L −1 at −15 • C, 0.22 and 0.16 L −1 at −20 • C, and 0.99 and 1.1 L −1 at −25 • C, respectively.These findings are consistent with previous measurements at similar locations.For example, Arctic measurements of Bigg (1996) for particles < 10 µm and Fountain and Ohtake (1985) using filter samples found mean INP concentrations of 0.01 L −1 at −15 • C and 0.13 L −1 at −20 • C, respectively, with both deposition and condensation modes likely possible, and Prenni et al. (2007) measured an average INP number concentration for particles with and aerodynamic diameter < 1.5 µm of approximately 0.33 L −1 between −8 and −28 • C with deposition, condensation, and immersion modes likely possible.At high-elevation sites, Bowers et al. (2009) at Mt. Werner in Colorado and Conen et al. (2012) at the research station Jungfraujoch in Switzerland measured mean immersionmode INP number concentrations of approximately 0.02 L −1 at −10 and −12 • C, respectively, using filter samples of 0.2-0.3µm pore size.High-elevation sites can receive large quantities of dust, which can act as efficient INPs at lower temperatures (Chou et al., 2011), but this was unlikely during our measurement period based on the low INP number concentrations.
INP number concentrations at Amphitrite Point were 0.23, 0.94, and 2.15 L −1 at droplet freezing temperatures of −15, −20, and −25 • C, respectively.Despite the predominance of marine air masses being sampled, the major source of INPs at Amphitrite Point during the study period was likely biological particles from local vegetation (Mason et al., 2015b).Similar values were measured in the Labrador Sea, where INP number concentrations at −15, −20, and −25 • C were 0.38, 1.3, and 2.8 L −1 , respectively.These concentrations are consistent with previous measurements within the marine boundary layer in regions influenced by air flow off of nearby coasts, for instance those of Schnell (1977) in the immersion freezing mode for sizes > 0.45 µm off the coast of Nova Scotia, roughly 1100-1500 km southwest of our sampling site in the Labrador Sea, and Rosinski et al. (1995) over the East China Sea in the deposition and condensation modes for sizes > 0.2 µm.However, INP number concentrations found during marine studies can vary by several orders of magnitude with changing location as summarized by Burrows et al. (2013).
The highest concentrations of INPs at a freezing temperature of −25 • C were found at the Colby sites, where the average number concentration was 8.9 L −1 .Aerosol sampling was conducted adjacent to soya and sorghum fields during and following periods of combine operation.This high concentration of INPs is consistent with previous work of Garcia et al. (2012) that showed elevated concentrations of INPs downwind of corn fields during combine harvesting, and Bowers et al. (2011) who found greater INP concentrations in air above cropland than above suburban or forest sites.

INP size distributions
Figure 3a shows the relative contribution of supermicron aerosol particles to the total measured INP population.Note that the same particle size range was not investigated at all locations with particles in the range of 0.10-0.18µm not being measured at Whistler Mountain or Amphitrite Point, and no uncertainty is reported for the Labrador Sea measurement as only a single sample was available.Averaging over all sampling locations with a 1 SD uncertainty, 91 ± 9, 79 ± 17, and 63 ± 21 % of INPs had an aerodynamic diameter > 1 µm at ice activation temperatures of −15, −20, and −25 • C, re- Figure 3b shows the fraction of INPs that are in the coarse mode, calculated by assuming that half of the INPs found in the 1.8-3.2µm MOUDI size cut were larger than 2.5 µm (i.e., INPs are uniformly distributed over this size range).Measurements of the total particle size distribution were not available at all locations to test this assumption.Furthermore, it is not known whether the INP size distribution follows the total particle size distribution a priori.Averaging over all sampling locations, the percentage of INPs in the coarse mode was 62 ± 20, 55 ± 18, and 42 ± 17 % ( 1 INP size distributions are further explored in Fig. 5, where the fraction of the measured INP number concentration found in each MOUDI size bin is shown.Colors on the red end of the scale illustrate that a large fraction of the INPs measured at a particular location belong to that particle size bin.Total INP number concentrations as a function of temperature are given in Fig. S1 and histograms of the INP size distributions are given in Figs.S2-S8. Figure 5a shows that most INPs active at −15 • C were 1-10 µm in size.In particular, when averaged over all locations, 72 % of the INPs active at −15 • C were between 1.8 and 5.6 µm.Furthermore, the major mode (i.e., the global maximum in a size distribution) was always larger than 1.8 µm.At lower freezing temperatures the INP size distributions broadened with increased contributions from smaller aerosol particles, evident by the more uniform intensity of Fig. 5b and c.By −25 • C, six of the seven locations had a submicron INP mode, and at Alert and Whistler Mountain this was the major mode.A general broadening of the INP size distribution with decreasing temperature would be expected if there were an increase in the number of particle types exhibiting ice activity with decreasing activation temperature.
Several previous studies have conducted size-resolved INP measurements.In most of these studies, the fraction of INPs larger than a given particle size was not reported, and in some cases the temperatures studied were different than in the current study.To better compare our data with these previous studies, we have used the literature data to calculate the fraction of INPs larger than either 1, 1.2, or 2.5 µm at temperatures as close as possible to the freezing temperatures we used.Details of the calculations are presented in the Supplement, and the results of the calculations are summarized in Table 2.
Table 2 shows that in six out of the nine previous studies, a large fraction (16-100 %) of the INPs was found to be supermicron in size, consistent with the current study.Here the work of Rosinski et al. (1986) is considered as two separate studies given the change in the investigated mode of ice nucleation.Although the condensation and immersion freezing  2007) is based on electrical mobility and optical measurements.In all other studies, particle size is given as the aerodynamic diameter.measurements reported in Rosinski et al. (1986) appear contradictory, it is important to note that the two freezing results for Rosinski et al. (1986) shown in Table 2 correspond to different temperature ranges (compare −5 to −6 • C for condensation and −10.8 • C for immersion freezing).One pos-sibility is that small particles dominated the INP population at the warmest temperatures while larger particles dominated the INP population at the colder temperatures investigated by Rosinski et al. (1986).
The ice nucleation efficiency of particles as a function of size have also been investigated in laboratory experiments (e.g., Lüönd et al., 2010;Archuleta et al., 2005;Welti et al., 2009).In general this work has shown that ice nucleation efficiency increases as particle size increases.Furthermore, several studies have investigated correlations between the concentrations of INPs and aerosol particles above a certain size (Richardson et al., 2007;DeMott et al., 2010;Chou et al., 2011;Field et al., 2012;Huffman et al., 2013;Prenni et al., 2013;Tobo et al., 2013;Ardon-Dryer and Levin, 2014;Jiang et al., 2014Jiang et al., , 2015)).For example, using data from a variety of field measurements DeMott et al. ( 2010) observed a correlation between INP concentrations and the concentration of aerosol particles > 0.5 µm, and Ardon-Dryer and Levin (2014) found that INP concentrations in Israel were better correlated to the concentration of aerosol particles 2.5-10 µm in size than those < 2.5 µm.These results are also consistent with INPs being relatively large in size.(2) at the lowest temperature analyzed, −25 • C, 63 ± 21 and 42 ± 17 % of INPs across all locations remained in the supermicron regime and coarse mode, respectively; (3) at −15 • C, the median INP size was relatively large at all locations, varying from 2.1 µm at Saclay to 4.7 µm at Colby with an average of 3.2 ± 0.9 µm; and (4) at −25 • C, the median size INP varied from 0.83 µm at Alert to 3.1 µm above the Labrador Sea with an average of 1.9 ± 1.0 µm.
Our measurements indicate that, when averaged over all studied locations and temperatures, 78 ± 19 and 53 ± 20 % of immersion-mode INPs may be missed if either supermicron particles or coarse-mode particles are not sampled at ground sites.As noted in Sect. 1, some instrumentation for measuring ambient INP number concentrations restricts the upper range of sampled aerosol particles.The data presented may be useful for estimating the fraction of INPs not measured with these instruments at ground sites and in different environments.
All measurements used in this study were conducted at ground level and, apart from those at Whistler Mountain, were also close to sea level.As the large contribution of supermicron and coarse-mode INPs to the overall INP population noted here may not necessarily hold for higher altitudes, additional size-resolved INP measurements as a function of altitude are needed.In obtaining such data, careful consideration will be needed toward sampling issues with aerosol inlets and transfer through sample lines on aircraft platforms.It can be anticipated that supermicron particle transfer will be strongly restricted without special inlets or care, and this may especially impact INP sampling.
A caveat to this study is that our measurements were confined to aerosol particle sizes greater than either 0.10 or 0.18 µm.If there were a significant contribution from INPs of smaller sizes (Vali, 1966;Schnell and Vali, 1973;Pummer et al., 2012;Augustin et al., 2013;Fröhlich-Nowoisky et al., 2015;O'Sullivan et al., 2015;Tong et al., 2015;Wilson et al., 2015), and these smaller sizes did not coagulate or get scavenged by larger particles, the values presented here would represent upper limits to the contribution of supermicron and coarse-mode particles to the total INP population.Additional studies exploring the relative atmospheric abundance of INPs < 0.10 µm are necessary (Hader et al., 2014).Future studies of the size distribution of INPs should also include measurements of particle mixing state to determine whether particles are internally or externally mixed at the locations where the size distribution of INPs are being measured.
The Supplement related to this article is available online at doi:10.5194/acp-16-1637-2016-supplement.
Aerosol particle collection at this alpine site occurred between 30 March and 23 April 2014.The louvered TSP inlet was located approximately 10 m from a chairlift operating station.Although there are no continuous combustion sources at the site, sampled air may have been influenced by engine exhaust for short periods of time due to nearby snow-1640 R. H. Mason et al.: Size-resolved measurements of ice-nucleating particles

Figure 2 .
Figure 2. Mean INP number concentrations at droplet freezing temperatures of −15 • C (dark gray), −20 • C (intermediate gray), and −25 • C (light gray).Uncertainty is given as the standard error of the mean, assuming a normal distribution.As only one sample was available from the Labrador Sea, no uncertainty is reported.

Figure 3 .
Figure 3.The mean fraction of INPs larger than (a) 1 µm and (b) 2.5 µm.Uncertainty is the standard error of the mean, assuming a normal distribution.Shading in the histogram corresponds to INP activation temperature: −15 • C is dark gray, −20 • C is an intermediate gray, and −25 • C is light gray.As 2.5 µm does not align with the size cut of a MOUDI stage, the fraction of INPs larger than 2.5 µm was found by assuming that number concentration of INPs 1.8-3.2µm in size was uniformly distributed over that size range.As only one sample was available from the Labrador Sea, no uncertainty is reported.

Figure 4 .
Figure 4.The median size of INPs at ice-activation temperatures of −15 • C (green), −20 • C (blue), and −25 • C (red) when averaged over all analyzed samples.Upper and lower uncertainties are the 75th and 25th percentiles, respectively.
SD) at ice activation temperatures of −15, −20, and −25 • C, respectively.The percentage of INPs in the coarse mode was found to range from 38 % at Saclay to 91 % in Colby at −15 • C, from 26 % at Whistler Mountain to 73 % in Colby at −20 • C, and from 20 % at Alert to 64 % at the Labrador Sea at −25 • C. Despite great diversity in the studied locations, each had a significant contribution from coarse-mode particles to the measured INP population.The median sizes of INPs at ice activation temperatures of −15, −20, and −25 • C are shown in Fig. 4 with the 25th and 75th percentile values.At −15 • C, the median INP size is relatively large at all locations, varying from 2.1 µm at Saclay to 4.7 µm at Colby with an average of 3.2 ± 0.9 µm (1 SD).As droplet freezing temperature decreased, the median INP size also decreased, with the exception of samples from the Labrador Sea and the UBC campus.At −25 • C, the median size of INPs varied from 0.83 µm at Alert to 3.1 µm at theLabrador Sea site with an average of 1.9 ± 1.0 µm.The median size of the INPs was > 1 µm in all cases with the exception of the Alert and Whistler Mountain sites at a freezing temperature of −25 • C. Alert, in the polar tundra at high latitude, and Whistler Mountain, at high elevation and periodically in the free troposphere, are remote with fewer local sources of aerosols.In Fig.4, the difference between the 75th and 25th percentile sizes is relatively small at all locations at −15 • C. A narrow INP size distribution at −15 • C is consistent with a single type or class of particles dominating freezing at this temperature.With decreasing temperature, the interquartile range significantly increased: the 75th and 25th percentile INP sizes decreased by an average factor of 1.2 and 3.7, respectively, between −15 and −25 • C, corresponding to a 63 % increase in the average interquartile range.

Figure 5 .
Figure 5. Fractional INP concentrations as a function of aerosol particle size, location, and activation temperature: (a) −15 • C; (b) −20 • C; and (c)−25 • C. The color bar indicates the fraction of INPs measured in each particle size bin.Aerosol particle sizes correspond to the 50 % cutoff aerodynamic diameters of the MOUDI stages(Marple et al., 1991).Missing sizes for the Whistler Mountain and Amphitrite Point sites are uncolored.

4
Summary and conclusions INP number concentrations in the immersion mode as a function of size and droplet freezing temperature were determined at six locations across North America and one in Europe.INP number concentrations varied by as much as an order of magnitude between locations, and were generally found to be lowest at the remote sites of Alert and Whistler Mountain and highest at the agricultural sites of Colby and the suburban sites of Saclay and the UBC campus, consistent with previous studies.Several key findings indicate the potential importance of large INPs at ground level: (1) 91 ± 9 and 62 ± 20 % of INPs measured at −15 • C across all locations are supermicron or in the coarse mode, respectively;

Table 1 .
(Marple et al., 1991)sed in this study and conditions during sampling.Aerodynamic diameter based on the 50 % cutoff of the MOUDI(Marple et al., 1991). a
Vali (1966)ata at the temperatures of this study (−15, −20, and −25 • C) when available, otherwise the next closest temperature was used.bTSP=total suspended particulate.cParticlesizereported forVali (1966)is based on filter pore size, and particle size reported inMertes et al. (